Exporting results
Exporting tables
The project files of MARADONER are not meant to be read by other software. The export command turns them into plain tables:
> maradoner export MyProject results
Here, results is the output folder. All tables in it are tab-separated, and the first column holds row names. The command runs after predict.
Besides the tables, the folder gets a README.md file. It describes every file found in the folder and every column of it, with respect to the options the tables were made with. The other commands writing into the same folder (contrast, anova, grn and so on) add their files to it. This page of the tutorial is therefore a guide to the folder, and the generated file is the reference.
What is in the folder?
| Folder | Content | Details |
|---|---|---|
| activities | Deviations of motif activities from their means in each group, their standard deviations and z-scores | Motif activities |
| params | Parameter estimates with standard errors | Below |
| tests | Tests for each motif | Tests |
| fov | Goodness-of-fit statistics, present if gof was run before the export |
Goodness of fit |
Motifs appear in the tables of activities and in most of the tests only if predict kept them.
Parameter estimates
| File | Rows | Content |
|---|---|---|
| motif_means.tsv | Motifs | The mean activity of each motif (mean) and its standard error (std) |
| motif_variances.tsv | Motifs | The variance of activity deviations per motif (tau) and its standard error (std) |
| filtered_motif_variances.tsv | Motifs | The motifs left out by predict for a negligible variance |
| group_variances.tsv | Groups | The error variance (sigma) and the multiplier of activity variances (nu) with their standard errors |
| promoter_means.tsv | Promoters | Promoter-wise means |
| sample_means.tsv | Samples | Sample-wise means |
| covariates/intercept.tsv | Motifs | The mean activity once more, with a two-sided test against zero: effect, std, z, pvalue, fdr |
| covariates/{covariate}.tsv | Motifs | For a project with covariates: the change of motif activity per unit of the covariate, with the same columns |
| factors.tsv | Samples | For a fit with --activity-lowrank: the factors, a column per factor. error_factors.tsv is the same for --error-lowrank |
| family_variances.tsv | Families | For a fit with --motif-families: the variance shared by the motifs of each family (omega), its standard error (std) and the number of motifs (motifs). The tau of a motif is then the variance of its individual part |
| family_factors.tsv | Families | For a fit with --motif-families and --motif-lowrank: the factors of the covariance between families, a column per factor. Without families the factors are in motif_factors.tsv, a row per motif |
| family_correlations.tsv | Families | For a fit with --motif-lowrank and families: the correlations between families implied by $Ω + L L^T$ |
| activity_correlations.tsv | Motifs | For a fit with --motif-families or --motif-lowrank: the correlations between the activities of motifs implied by the fitted covariance between motifs |
| correlations/ | Correlations between the estimates of a kind, e.g. between the mean activities of different motifs |
The variance of the activity of motif $k$ in group $g$ is the product $τ_k ν_g$. A product does not change if one factor is multiplied and the other is divided by the same number, so the scale of nu is arbitrary. MARADONER settles it by fixing the nu of one group. The standard error of that nu is left empty.
Optional tables
Some tables are large and are written on request only:
| Name | Description |
|---|---|
--log-counts |
The expression predicted by the full model is written to the log_counts folder: one column per group, or one per sample with --no-log-counts-grouped |
--loadings-product |
For each group, a table of promoters by motifs is written to the loadings-product folder: the loading times the activity (the mean plus the deviation), i.e. the contribution of a motif to the expression of a promoter. The tables are in the HDF format, and --no-lp-hdf switches to plain text. With --no-lp-intercepts the mean activity is left out |
A few more options change the content of the regular tables:
| Name | Description |
|---|---|
--weighted-zscore |
The overall column of z_score.tsv becomes the absolute z-score of the mean of the group activities weighted by inverse variances |
--zero-margin <float> |
Turns the zero column of the sign test into an equivalence test |
--anova-mode <str> |
Chooses the tail of the ANOVA statistic |
